Drones Monitor Crowds for FIFA World Cup 2034 with Label-Free Adaptation

A new protocol for drone-based crowd monitoring has been developed and rigorously validated, specifically designed for large-scale events such as the upcoming FIFA World Cup in Saudi Arabia. This system aims to provide accurate crowd counts from aerial footage, even when the data differs significantly from the training set, and crucially, to detect dangerous crowd inflows before crushes occur. The core innovation lies in its label-free adaptation capabilities, which effectively mitigate errors caused by shifts in data distribution. Through extensive testing, including 525 controlled runs and various ablations, the method demonstrated substantial improvements in accuracy, recovering 31-49% of shift-induced errors. It also established a "severity law" to differentiate method performance and a "stability budget" for safe drone operation. Furthermore, the system successfully addressed dense-scene undercounting, a critical issue that could lead to under-reporting forming crushes. Its flux-based risk module accurately identified real congestion episodes in test clips. The research also provides a six-point deployment protocol, emphasizing unconditional adaptation with tail monitoring as a key policy for robust crowd management.